I have installed CUDA Toolkit 5.0 and nvidia-experimental-310 drivers. CUDA examples works fine with optirun
but every OpenCL app fails to run with error calling clGetPlatformIDs
function. I've tried examples from NVIDIA OpenCL SDK and also examples from PyOpenCL which I compiled from source and also using pip
, nothing works. Why it's not working?
I finally managed to get it work after creating
/etc/OpenCL/vendors/nvidia.icd
file andlibnvidia-opencl.so.1
inside. I also switched back tonvidia-current
driver before but I'm sure it still would work with experimental ones. I found a solution here.